diff --git a/Modules/SurfaceInterpolation/mitkCreateDistanceImageFromSurfaceFilter.cpp b/Modules/SurfaceInterpolation/mitkCreateDistanceImageFromSurfaceFilter.cpp index 8e8508388a..522698602d 100644 --- a/Modules/SurfaceInterpolation/mitkCreateDistanceImageFromSurfaceFilter.cpp +++ b/Modules/SurfaceInterpolation/mitkCreateDistanceImageFromSurfaceFilter.cpp @@ -1,629 +1,629 @@ /*=================================================================== The Medical Imaging Interaction Toolkit (MITK) Copyright (c) German Cancer Research Center, Division of Medical and Biological Informatics. All rights reserved. This software is distributed WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See LICENSE.txt or http://www.mitk.org for details. ===================================================================*/ #include "mitkCreateDistanceImageFromSurfaceFilter.h" mitk::CreateDistanceImageFromSurfaceFilter::CreateDistanceImageFromSurfaceFilter() { m_DistanceImageVolume = 50000; this->m_UseProgressBar = false; this->m_ProgressStepSize = 5; mitk::Image::Pointer output = mitk::Image::New(); this->SetNthOutput(0, output.GetPointer()); } mitk::CreateDistanceImageFromSurfaceFilter::~CreateDistanceImageFromSurfaceFilter() { } void mitk::CreateDistanceImageFromSurfaceFilter::GenerateData() { //First of all we have to build the equation-system from the existing contour-edge-points this->CreateSolutionMatrixAndFunctionValues(); //Then we solve the equation-system via QR - decomposition. The interpolation weights are obtained in that way vnl_qr solver (m_SolutionMatrix); m_Weights = solver.solve(m_FunctionValues); //Setting progressbar if (this->m_UseProgressBar) mitk::ProgressBar::GetInstance()->Progress(2); //The last step is to create the distance map with the interpolated distance function this->CreateDistanceImage(); m_Centers.clear(); m_FunctionValues.clear(); m_Normals.clear(); m_Weights.clear(); m_SolutionMatrix.clear(); //Setting progressbar if (this->m_UseProgressBar) mitk::ProgressBar::GetInstance()->Progress(3); } void mitk::CreateDistanceImageFromSurfaceFilter::CreateSolutionMatrixAndFunctionValues() { unsigned int numberOfInputs = this->GetNumberOfIndexedInputs(); if (numberOfInputs == 0) { MITK_ERROR << "mitk::CreateDistanceImageFromSurfaceFilter: No input available. Please set an input!" << std::endl; itkExceptionMacro("mitk::CreateDistanceImageFromSurfaceFilter: No input available. Please set an input!"); return; } //First of all we have to extract the nomals and the surface points. //Duplicated points can be eliminated Surface* currentSurface; vtkSmartPointer polyData; vtkSmartPointer currentCellNormals; vtkSmartPointer existingPolys; vtkSmartPointer existingPoints; double p[3]; PointType currentPoint; PointType normal; for (unsigned int i = 0; i < numberOfInputs; i++) { currentSurface = const_cast( this->GetInput(i) ); polyData = currentSurface->GetVtkPolyData(); if (polyData->GetNumberOfPolys() == 0) { MITK_INFO << "mitk::CreateDistanceImageFromSurfaceFilter: No input-polygons available. Please be sure the input surface consists of polygons!" << std::endl; } currentCellNormals = vtkDoubleArray::SafeDownCast(polyData->GetCellData()->GetNormals()); existingPolys = polyData->GetPolys(); existingPoints = polyData->GetPoints(); existingPolys->InitTraversal(); vtkIdType* cell (NULL); vtkIdType cellSize (0); for( existingPolys->InitTraversal(); existingPolys->GetNextCell(cellSize, cell);) { for ( unsigned int j = 0; j < cellSize; j++ ) { existingPoints->GetPoint(cell[j], p); currentPoint.copy_in(p); int count = std::count(m_Centers.begin() ,m_Centers.end(),currentPoint); if (count == 0) { double currentNormal[3]; currentCellNormals->GetTuple(cell[j], currentNormal); normal.copy_in(currentNormal); m_Normals.push_back(normal); m_Centers.push_back(currentPoint); } }//end for all points }//end for all cells }//end for all outputs //For we can now calculate the exact size of the centers we initialize the data structures unsigned int numberOfCenters = m_Centers.size(); m_Centers.reserve(numberOfCenters*3); m_FunctionValues.set_size(numberOfCenters*3); m_FunctionValues.fill(0); //Create inner points for (unsigned int i = 0; i < numberOfCenters; i++) { currentPoint = m_Centers.at(i); normal = m_Normals.at(i); currentPoint[0] = currentPoint[0] - normal[0]; currentPoint[1] = currentPoint[1] - normal[1]; currentPoint[2] = currentPoint[2] - normal[2]; m_Centers.push_back(currentPoint); m_FunctionValues.put(numberOfCenters+i, -1); } //Create outer points for (unsigned int i = 0; i < numberOfCenters; i++) { currentPoint = m_Centers.at(i); normal = m_Normals.at(i); currentPoint[0] = currentPoint[0] + normal[0]; currentPoint[1] = currentPoint[1] + normal[1]; currentPoint[2] = currentPoint[2] + normal[2]; m_Centers.push_back(currentPoint); m_FunctionValues.put(numberOfCenters*2+i, 1); } //Now we have created all centers and all function values. Next step is to create the solution matrix numberOfCenters = m_Centers.size(); m_SolutionMatrix.set_size(numberOfCenters, numberOfCenters); m_Weights.set_size(numberOfCenters); PointType p1; PointType p2; double norm; for (unsigned int i = 0; i < numberOfCenters; i++) { for (unsigned int j = 0; j < numberOfCenters; j++) { //Calculate the RBF value. Currently using Phi(r) = r with r is the euclidian distance between two points p1 = m_Centers.at(i); p2 = m_Centers.at(j); p1 = p1 - p2; norm = p1.two_norm(); m_SolutionMatrix(i,j) = norm; } } } void mitk::CreateDistanceImageFromSurfaceFilter::CreateDistanceImage() { DistanceImageType::Pointer distanceImg = DistanceImageType::New(); // Determine the bounds of the input points in index- and world-coordinates DistanceImageType::PointType minPointInWorldCoordinates, maxPointInWorldCoordinates; DistanceImageType::IndexType minPointInIndexCoordinates, maxPointInIndexCoordinates; DetermineBounds( minPointInWorldCoordinates, maxPointInWorldCoordinates, minPointInIndexCoordinates, maxPointInIndexCoordinates ); // Calculate the extent of the region that contains all given points in MM. // To do this, we take the difference between the maximal and minimal // index-coordinates (must not be less than 1) and multiply it with the // spacing of the reference-image. Vector3D extentMM; for (unsigned int dim = 0; dim < 3; ++dim) { extentMM[dim] = (int) ( (std::max( std::abs(maxPointInIndexCoordinates[dim] - minPointInIndexCoordinates[dim]), (DistanceImageType::IndexType::IndexValueType) 1 ) + 1.0) // (max-index - min-index)+1 because the pixels between index 3 and 5 cover 2+1=3 pixels (pixel 3,4, and 5) * m_ReferenceImage->GetSpacing()[dim] ) + 1; // (int) ((...) + 1) -> we round up to the next BIGGER int value } /* * Now create an empty distance image. The create image will always have the same sizeOfRegion, independent from * the original image (e.g. always consists of 500000 pixels) and will have an isotropic spacing. * The spacing is calculated like the following: * The image's volume = 500000 Pixels = extentX*spacing*extentY*spacing*extentZ*spacing * So the spacing is: spacing = ( 500000 / extentX*extentY*extentZ )^(1/3) */ double basis = (extentMM[0]*extentMM[1]*extentMM[2]) / m_DistanceImageVolume; double exponent = 1.0/3.0; double distImgSpacing = pow(basis, exponent); int tempSpacing = (distImgSpacing+0.05)*10; m_DistanceImageSpacing = (double)tempSpacing/10.0; // calculate the number of pixels of the distance image for each direction unsigned int numberOfXPixel = extentMM[0] / m_DistanceImageSpacing; unsigned int numberOfYPixel = extentMM[1] / m_DistanceImageSpacing; unsigned int numberOfZPixel = extentMM[2] / m_DistanceImageSpacing; // We increase the sizeOfRegion by 4 as we decrease the origin by 2 later. // This expansion of the region is necessary to achieve a complete // interpolation. DistanceImageType::SizeType sizeOfRegion; sizeOfRegion[0] = numberOfXPixel + 4; sizeOfRegion[1] = numberOfYPixel + 4; sizeOfRegion[2] = numberOfZPixel + 4; // The region starts at index 0,0,0 DistanceImageType::IndexType initialOriginAsIndex; initialOriginAsIndex.Fill(0); DistanceImageType::PointType originAsWorld = minPointInWorldCoordinates; DistanceImageType::RegionType lpRegion; lpRegion.SetSize(sizeOfRegion); lpRegion.SetIndex(initialOriginAsIndex); // We initialize the itk::Image with // * origin and direction to have it correctly placed and rotated in the world // * the largest possible region to set the extent to be calculated // * the isotropic spacing that we have calculated above distanceImg->SetOrigin( originAsWorld ); distanceImg->SetDirection( m_ReferenceImage->GetDirection() ); distanceImg->SetRegions( lpRegion ); distanceImg->SetSpacing( m_DistanceImageSpacing ); distanceImg->Allocate(); //First of all the image is initialized with the value 10 for each pixel distanceImg->FillBuffer(10); // Now we move the origin of the distanceImage 2 index-Coordinates // in all directions DistanceImageType::IndexType originAsIndex; distanceImg->TransformPhysicalPointToIndex( originAsWorld, originAsIndex ); originAsIndex[0] -= 2; originAsIndex[1] -= 2; originAsIndex[2] -= 2; distanceImg->TransformIndexToPhysicalPoint( originAsIndex, originAsWorld ); distanceImg->SetOrigin( originAsWorld ); /* * Now we must calculate the distance for each pixel. But instead of calculating the distance value * for all of the image's pixels we proceed similar to the region growing algorithm: * * 1. Take the first pixel from the narrowband_point_list and calculate the distance for each neighbor (6er) * 2. If the current index's distance value is below a certain threshold push it into the list * 3. Next iteration take the next index from the list and originAsIndex with 1. again * * This is done until the narrowband_point_list is empty. */ std::queue narrowbandPoints; PointType currentPoint = m_Centers.at(0); double distance = this->CalculateDistanceValue(currentPoint); // create itk::Point from vnl_vector DistanceImageType::PointType currentPointAsPoint; currentPointAsPoint[0] = currentPoint[0]; currentPointAsPoint[1] = currentPoint[1]; currentPointAsPoint[2] = currentPoint[2]; // Transform the input point in world-coordinates to index-coordinates DistanceImageType::IndexType currentIndex; distanceImg->TransformPhysicalPointToIndex( currentPointAsPoint, currentIndex ); assert( lpRegion.IsInside(currentIndex) ); // we are quite certain this should hold narrowbandPoints.push(currentIndex); distanceImg->SetPixel(currentIndex, distance); NeighborhoodImageIterator::RadiusType radius; radius.Fill(1); NeighborhoodImageIterator nIt(radius, distanceImg, distanceImg->GetLargestPossibleRegion()); unsigned int relativeNbIdx[] = {4, 10, 12, 14, 16, 22}; bool isInBounds = false; while ( !narrowbandPoints.empty() ) { nIt.SetLocation(narrowbandPoints.front()); narrowbandPoints.pop(); unsigned int* relativeNb = &relativeNbIdx[0]; for (int i = 0; i < 6; i++) { nIt.GetPixel(*relativeNb, isInBounds); if( isInBounds && nIt.GetPixel(*relativeNb) == 10) { currentIndex = nIt.GetIndex(*relativeNb); // Transform the currently checked point from index-coordinates to // world-coordinates distanceImg->TransformIndexToPhysicalPoint( currentIndex, currentPointAsPoint ); // create a vnl_vector currentPoint[0] = currentPointAsPoint[0]; currentPoint[1] = currentPointAsPoint[1]; currentPoint[2] = currentPointAsPoint[2]; // and check the distance distance = this->CalculateDistanceValue(currentPoint); if ( abs(distance) <= m_DistanceImageSpacing ) { nIt.SetPixel(*relativeNb, distance); narrowbandPoints.push(currentIndex); } } relativeNb++; } } ImageIterator imgRegionIterator (distanceImg, distanceImg->GetLargestPossibleRegion()); imgRegionIterator.GoToBegin(); double prevPixelVal = 1; unsigned int _size[3] = { (unsigned int)(sizeOfRegion[0] - 1), (unsigned int)(sizeOfRegion[1] - 1), (unsigned int)(sizeOfRegion[2] - 1) }; - double center [3] = {_size[0]/2, _size[1]/2, _size[2]/2}; + double center [3] = {_size[0]/2.0, _size[1]/2.0, _size[2]/2.0}; MITK_INFO<<"Size: ["<<_size[0]<<","<<_size[1]<<","<<_size[2]<<"] Center: ["<GetOutput(); // Cast the created distance-Image from itk::Image to the mitk::Image // that is our output. CastToMitkImage(distanceImg, resultImage); } void mitk::CreateDistanceImageFromSurfaceFilter::FillImageRegion(DistanceImageType::RegionType reqRegion, DistanceImageType::PixelType pixelValue, DistanceImageType::Pointer image) { image->SetRequestedRegion(reqRegion); ImageIterator it (image, image->GetRequestedRegion()); while (!it.IsAtEnd()) { it.Set(pixelValue); ++it; } } double mitk::CreateDistanceImageFromSurfaceFilter::CalculateDistanceValue(PointType p) { double distanceValue (0); PointType p1; PointType p2; double norm; CenterList::iterator centerIter; InterpolationWeights::iterator weightsIter; for ( centerIter=m_Centers.begin(), weightsIter=m_Weights.begin(); centerIter!=m_Centers.end() && weightsIter!=m_Weights.end(); centerIter++, weightsIter++ ) { p1 = *centerIter; p2 = p-p1; norm = p2.two_norm(); distanceValue = distanceValue + norm* (*weightsIter); } return distanceValue; } void mitk::CreateDistanceImageFromSurfaceFilter::GenerateOutputInformation() { } void mitk::CreateDistanceImageFromSurfaceFilter::PrintEquationSystem() { std::ofstream esfile; esfile.open("C:/Users/fetzer/Desktop/equationSystem/es.txt"); esfile<<"Nummber of rows: "<SetInput( 0, const_cast( surface ) ); } void mitk::CreateDistanceImageFromSurfaceFilter::SetInput( unsigned int idx, const mitk::Surface* surface ) { if ( this->GetInput(idx) != surface ) { this->SetNthInput( idx, const_cast( surface ) ); this->Modified(); } } const mitk::Surface* mitk::CreateDistanceImageFromSurfaceFilter::GetInput() { if (this->GetNumberOfIndexedInputs() < 1) return NULL; return static_cast(this->ProcessObject::GetInput(0)); } const mitk::Surface* mitk::CreateDistanceImageFromSurfaceFilter::GetInput( unsigned int idx) { if (this->GetNumberOfIndexedInputs() < 1) return NULL; return static_cast(this->ProcessObject::GetInput(idx)); } void mitk::CreateDistanceImageFromSurfaceFilter::RemoveInputs(mitk::Surface* input) { DataObjectPointerArraySizeType nb = this->GetNumberOfIndexedInputs(); for(DataObjectPointerArraySizeType i = 0; i < nb; i++) { if( this->GetInput(i) == input ) { this->RemoveInput(i); return; } } } void mitk::CreateDistanceImageFromSurfaceFilter::Reset() { for (unsigned int i = 0; i < this->GetNumberOfIndexedInputs(); i++) { this->PopBackInput(); } this->SetNumberOfIndexedInputs(0); this->SetNumberOfIndexedOutputs(1); mitk::Image::Pointer output = mitk::Image::New(); this->SetNthOutput(0, output.GetPointer()); } void mitk::CreateDistanceImageFromSurfaceFilter::SetUseProgressBar(bool status) { this->m_UseProgressBar = status; } void mitk::CreateDistanceImageFromSurfaceFilter::SetProgressStepSize(unsigned int stepSize) { this->m_ProgressStepSize = stepSize; } void mitk::CreateDistanceImageFromSurfaceFilter::SetReferenceImage( itk::ImageBase<3>::Pointer referenceImage ) { m_ReferenceImage = referenceImage; } void mitk::CreateDistanceImageFromSurfaceFilter::DetermineBounds( DistanceImageType::PointType &minPointInWorldCoordinates, DistanceImageType::PointType &maxPointInWorldCoordinates, DistanceImageType::IndexType &minPointInIndexCoordinates, DistanceImageType::IndexType &maxPointInIndexCoordinates ) { PointType firstCenter = m_Centers.at(0); DistanceImageType::PointType tmpPoint; tmpPoint[0] = firstCenter[0]; tmpPoint[1] = firstCenter[1]; tmpPoint[2] = firstCenter[2]; // transform the first point from world-coordinates to index-coordinates DistanceImageType::IndexType tmpIndex; m_ReferenceImage->TransformPhysicalPointToIndex( tmpPoint, tmpIndex ); // initialize the variables with this first point int xmin = tmpIndex[0]; int ymin = tmpIndex[1]; int zmin = tmpIndex[2]; int xmax = tmpIndex[0]; int ymax = tmpIndex[1]; int zmax = tmpIndex[2]; // iterate over the rest of the points CenterList::iterator centerIter = m_Centers.begin(); for ( ++centerIter; centerIter!=m_Centers.end(); centerIter++) { tmpPoint[0] = (*centerIter)[0]; tmpPoint[1] = (*centerIter)[1]; tmpPoint[2] = (*centerIter)[2]; // transform each point from world-coordinates to index-coordinates m_ReferenceImage->TransformPhysicalPointToIndex( tmpPoint, tmpIndex ); // and set the variables accordingly to find the minimum // and maximum in all directions in index-coordinates if (xmin > tmpIndex[0]) { xmin = tmpIndex[0]; } if (ymin > tmpIndex[1]) { ymin = tmpIndex[1]; } if (zmin > tmpIndex[2]) { zmin = tmpIndex[2]; } if (xmax < tmpIndex[0]) { xmax = tmpIndex[0]; } if (ymax < tmpIndex[1]) { ymax = tmpIndex[1]; } if (zmax < tmpIndex[2]) { zmax = tmpIndex[2]; } } // put the found coordinates into Index-Points minPointInIndexCoordinates[0] = xmin; minPointInIndexCoordinates[1] = ymin; minPointInIndexCoordinates[2] = zmin; maxPointInIndexCoordinates[0] = xmax; maxPointInIndexCoordinates[1] = ymax; maxPointInIndexCoordinates[2] = zmax; // and transform them into world-coordinates m_ReferenceImage->TransformIndexToPhysicalPoint( minPointInIndexCoordinates, minPointInWorldCoordinates ); m_ReferenceImage->TransformIndexToPhysicalPoint( maxPointInIndexCoordinates, maxPointInWorldCoordinates ); }